For almost a decade, I’ve had a recurring dream where I’m standing on the edges on an island and watching the missiles start to rise over the horizon. The dream always starts the same but how it progresses varies: sometimes we’re fighting from the beaches, others in a city I can’t quite recognize, sometimes the enemy has close air support, others we’re fighting hand to hand. But in all of those dreams the war never comes to an end. I always try something new to win the day, but I can never make it to the end. And that’s what attrition warfare is: waking every day into the same nightmare, clawing at the walls hoping to change just enough to break the fight for inches.

Every day, there’s new articles and pitches on how to bring autonomy to bear on the battlefield, specifically on how to deter and defeat an invasion of Taiwan. We’ve spent the better part of the last decade focused on deterrence and the early hours and days of World War III in the 1st Island Chain. These days, however, I find myself more and more obsessed with solving the problem of autonomous mass on the battlefield when a conflict with the PRC inevitably turns to attrition.
What is attrition?
Attrition at its simplest means wearing down your enemy until they’ve exhausted their resources and their ability or will to fight has collapsed. This strategy stands in contrast to maneuver warfare where the goal is to defeat your opponent through rapid action thereby breaking their cohesion and organization. Despite its difficulty, rarely does the military commander win praise for victory through attrition as brilliant. No one likes to plan for attrition. In the romantic view of warfare, maneuver is sexy whereas attrition is ugly and gluttonous. Wars, particularly long wars, are rarely just maneuver or attrition. War shifts between phases of attrition and maneuver as geography, logistics, mass, stupidity, and other contributing factors impact the battle lines. Very rarely does the enemy consent to your brilliant pre-war plan of masterful maneuver and an early knockout blow.
In the early 2020s, some of us made a bet: that deterrence by denial wasn’t enough. There wasn’t a fancy enough weapon or battle plan that could deter and win against a PLA invasion in a single swoop. We could not sink the PLAN in 72 hours as some claimed, we were not building systems and munitions at the rate we’d need to sustain even an effective opening salvo nor were we creating the supply and logistics chains necessary to enable such a plan. Ukraine reminded us that whatever our math for our missile magazines was; it wasn’t enough. Denial wasn’t enough: you’d have to exhaust the PLA. The conquest of Taiwan is an existential political problem for the CCP. Therefore, the sprint to 2027 or whatever target date you wanted to throw your darts at required duct tape and the kitchen sink. But that was just to buy us time, what you really needed was a whole of government approach that invested in a strategy of attrition. The PLA wanted the invasion to go as fast as possible (the fait accompli) so we had to slow it down and keep our foot on the gas. Building more missiles solved part of that problem: you build enough platforms and keep them distributed, you confound PLA targeting and slow down and limit their ability to wipe us out. But eventually we’d run out of missiles or our targeting would be disrupted by the PLA, and the PLA could build faster than we could…so all they’d need to do is wait us out and send a few boats across the Strait to the smoldering island of Taiwan. We needed mass, but the old measures of combat power couldn’t be generated fast enough. The only way to win was to change the rules.
Enter the warbot.
Why do we need autonomous mass?
In the Pacific, the US is outmatched in battlefield numbers, geography, production, and in some cases firepower. We sit at the edge of war with an enemy that can produce more missiles, ships, and people than we can ever hope. By producing relatively cheap, punchy autonomous systems which we put forward and pair with our existing exquisite capabilities we offset many of the advantages of the PRC. In other words, our approach has become “let the machines eat the first bullet.” For the last 5+ years, eating the first proverbial bullet has been the primary concern of a Taiwan fight: (how do we survive day 1 and bludgeon the enemy enough to make it to day 2?). You may know this better as “defeating a fait accompli against Taiwan.” Of course this solves one problem but leaves us poorly positioned to solve the next: what happens when the PLA doesn’t give up quickly?
Until recently, autonomous mass was a theoretical concept. The warbot a thing of science fiction and glossy PowerPoint slide decks (and for some, still is.) The remotely-piloted Predator Drone and similar systems of the GWOT were hardly a fit for a high end fight in the Pacific. We are building a new arsenal of all shapes and sizes to accompany the human half of the Joint Force. But simply owning and building these systems isn’t enough; you need to be able to deploy them, sustain them, and use them correctly. (You can read all of my works on combined minds doctrine and warbot development in these links.)
Where does autonomous mass fit in the attrition fight?
Not all autonomous mass is attritable, nor should it be. Sometimes autonomous systems are in fact exquisite and that’s ok. Something people often get wrong is assuming all warbots are attritable and that is a gross misunderstanding of both their battlefield application and the technology itself. When we talk about the warbot that can be everything from a major combat system like a ship or tank to a breadbox-sized drone. These all deliver different effects, with different quality of equipment, for different missions. You can’t just dump a fully autonomous formation in front of the enemy, let them eat the bullets, and assume that was the most effective use of millions of dollars of hardware. Sure, it’s better if the bot eats the bullet, but how that happens matters too. If the bot can survive and live to fight another day, even better. And while we’ve rapidly advanced in the last couple of years, the underlying technologies still face a lot of limitations. Warbots aren’t magic.
A bunch of quadrotor drones aren’t saving Taiwan by themselves, and the USS Iowa battleship can’t be turned into a robot tomorrow. The bigger the system, the more complex the compute and more capable the firepower (but also the longer the production line). As I’ve said before, this is a combined arms problem: you have to figure out how to mix and match your warbots to generate the best effects while being cognizant of the logistical tail behind you.
Similarly, this is roughly how the PLA talks about the warbot: smaller bots work from, with, and on behalf of larger bots (motherships).
For the US, autonomous mass in an attrition fight means reducing the risks posed by our current force structure (all volunteer force and smaller population, long-dev but high quality big ticket systems, and a lag in traditional defense production). If you can generate a lot of autonomous systems that can fight and harass the PLA every day without risk of catastrophic loss at a much lower cost, you can thereby lprioritze the exquisite systems for the big battles and critical moments (ie, enabling us to generate the mass we couldn’t otherwise in the face of PLA firepower.) This can be thought of in four principles:
1. When possible, the warbot should enter the kill box first.
When possible, the warbots should eat the bullet first.
The warbot should never be the single point of failure.
Maintenance for a warbot should never be more complex than its manned equivalent.
What are the tradeoffs and limitations of autonomous mass?
Like I said, warbots aren’t magic. War is a political endeavor and so the presence of people on the ground is the manifestation of the forcibly changed political reality, not a warbot. The war isn’t fought for the warbots, it’s fought for the people. So there is a difference between a couple hundred warbots stationed on an island vs a battalion of soldiers or Marines on the same island. Where the warbot matters most in the attritional battle is in contesting ground and forcing the enemy to expend additional resources. It takes time, resources, and a considerable amount of risk to put capital assets into battle, burn exquisite fires, and train and deploy people. The warbot buys down that risk when you’re talking about casualty numbers like we see in Ukraine or will see in the Pacific. But for all the bots you buy and deploy, you still have to host, feed, and maintain them. If you buy a lot of small, cheap systems then you get to have a lot of them and they can be easily replaced without much concern…but they also likely aren’t as reliable and possess limited capabilities in the categories of range, firepower, sensors, and compute. If you invest in super exquisite autonomy, that supply chain and cost likely comes at a big price (and commanders will be less willing to employ them for risk of losing them). For autonomous mass to work for the Joint Force, you have to find that balance between capability and numerical superiority. And let’s not forget: you have to get all of this to the fight, so how are you valuing the expense of precious logistics assets between the various classes of autonomous systems *and* between autonomous and manned systems.
What happens when the PRC scales autonomous mass?
In our pursuit of autonomous mass, we must understand that the advantage we gain will be forever fleeting. Welcome to the real arms race. It’s not about AGI, missiles, or ships. The real arms race of the Second Cold War is all about the warbot.
We don’t have a good picture into what the Chinese equivalent of the defense tech reformation looks like. There are things we can infer from clips of new drones emerging over the skies of a Chinese city or from public discussions around their doctrinal concept of “mothership warfare” but when and how they turn on the spigot of warbot production from their vast reservoir of industrial capacity is unclear. Doctrine and talking points are only as good as the war-ready combat formations to which they attach themselves. The thing that keeps me up at night (other than the nightmares) is what happens when the PRC goes all in on autonomous mass. (Yet another reason to choke their chip supply, btw.) The PRC’s industrial capacity and flexibility resembles ours in the leadup to WWII and they are far better postured to “turn on” their warbot assembly line than we are in a war of attrition.
How do we win?
When they finally decide to turn on the autonomous assembly lines for real, we have to be ready to out-iterate them. It’s not just about who gets version 1.0 to the field first, it’s about who can get every hardware and software update to the field first, every time. This may sound simply, but I assure you it’s not. There are three key factors to winning a war of autonomous attrition:
Software and hardware iteration: The ability to update sensors and code quickly and thoroughly without being dragged down by long lead times. Iterations of new technologies in Ukraine can measure in days.
Physical deployment: If we cannot get the gear to the field, then we cannot fight with the gear. Logistics, logistics, logistics.
Formation Integration: Again, autonomous attrition is not just about smashing our cheap toys with their cheap toys. C2, good staff work, and units trained in man-machine teaming make operations smoother and more effective. These successes will build upon one another, or they can break the chain if your network of warbots is all built on a broken API.
This is hardly the first time I’ve thought about autonomous attrition: fighting autonomous mass in a war of attrition in the Indo-Pacific is one of the key themes of my first novel, Ex Supra. In Ex Supra, the Americans forego weaponized AI after a backlash against domestic abuses (sound familiar?) and suffer the onslaught of a PLA that invested heavily in warbots. The common American vision of war is usually the highlight reel: the training montage, the missiles flying, the heroic charge into battle, the final assault on the enemy, the survivor walking into the sunrise. Our vision of war, both culturally and doctrinally, is expeditionary maneuver. The idea of staging and deploying thousands upon thousands of autonomous systems in mass, acting in concert with humans, day after day on the other side of the planet is a new one for us but hardly impossible.
The secret to winning a war of autonomous attrition isn’t to solely bet on autonomy, but to bet on man-machine teaming that amplifies the most potent effects of both. We cannot view humans as the weakest link in combat, nor can we afford to ignore how rapidly autonomy has improved for the warbot. Autonomous mass affords us the ability, as a less-populous, less-industrious nation, to survive the weight of attritive battle long enough to wear down the PLA and swing heavy with our advantages in expeditionary maneuver. If we want to win the 12-round title fight, we have to be able to take punches just as hard as we throw them. You have to go into the fight prepared for that knockdown, drag-out brawl in the ring. You can’t just plan for a one-round sucker punch. And we can’t wait until we run out of missiles and ships to start figuring out how to bring autonomous mass to bear 8, 15, 24 or 36 months into the war.
PS: I’ve got a new podcast: Second Breakfast w/ Jordan Schneider, Justin McIntosh, and Eric Robinson. It’s all about defense tech, policy, and warfare.
If you would like to read more about the future of US-China conflict, the failures of US foreign policy, or what happens next, check out my novel, EX SUPRA. It’s all about the world after the fall of Taiwan, an isolationist and hyper-partisan America, and World War III. It was nominated for a Prometheus Award for best science fiction novel and there’s a sequel in the works! Don’t forget to share and subscribe!
And that’s what attrition warfare is: waking into the same nightmare, clawing at the walls hoping to change enough to break the stalemate.